The verdict in three sentences
A construction firm that answers 120 public tenders a year ties up 2 to 3 days of bid engineer time per technical proposal, close to 300 days a year. A structured reference library and a generative AI that writes first drafts, always reviewed by a human, cut that time in half. In 2026 the market charges roughly EUR 12,000 to 30,000 for setup and under EUR 200 a month in API costs, an investment that pays back within a quarter for a mid-sized contractor.
Where the 2 to 3 days actually go
The technical proposal often carries 40 to 60% of the final score on works contracts. Yet much of the time spent is not expertise: finding the right project reference, rewriting the staffing and equipment section, adapting the health, safety and environment chapter already written twenty times. Here is the breakdown we see at mid-sized structural and fit-out contractors.
| Step | Time without AI | Time with AI and review | Who does it |
|---|---|---|---|
| Reading the tender documents and evaluation rules | 3 h | 1.5 h (automatic summary, human check) | Bid engineer |
| Selecting project references | 2.5 h | 0.5 h | Bid engineer |
| Staffing and equipment | 2 h | 0.5 h | Technical assistant |
| Site-specific method statement | 6 h | 4 h | Site manager |
| Health, safety, environment, waste | 3 h | 1 h | HSE manager |
| Programme and phasing | 2 h | 1.5 h | Site manager |
| Layout and final proofreading | 2 h | 1 h | Technical assistant |
| Total | 20.5 h, about 2.5 days | 10 h, about 1.25 days |
The method statement resists automation the most, by design: it is where the public buyer judges your grasp of the site, access constraints and phasing in an occupied building. The AI proposes a structure there, never a final answer.
What a serious setup includes
The heart of the project is the library, not the language model. You need to index 5 to 10 years of winning and losing proposals, project reference sheets (client, contract value, duration, completion certificate), CVs, certifications and HSE procedures. A search engine then retrieves the relevant passages for each evaluation criterion, and the model drafts a first version that cites its internal sources.
| Item | 2026 market estimate in Europe | Scope |
|---|---|---|
| Archive audit and structuring | EUR 2,500 to 6,000 | Sorting 200 to 600 proposals, standardised reference sheets |
| Indexing and document search | EUR 3,000 to 7,000 | Splitting by criterion, semantic search |
| First-draft generation interface | EUR 3,500 to 9,000 | One screen per criterion, cited sources, Word export |
| Assisted reading of tender documents | EUR 1,500 to 4,000 | Extracting criteria, weightings, required documents |
| Team training and support | EUR 1,500 to 4,000 | 2 to 4 workshops, review guidelines |
| Total setup | EUR 12,000 to 30,000 | Depends on archive volume |
| Model API and hosting | EUR 80 to 200 a month | For 10 proposals a month |
If Kolonell builds it, this tool falls under our custom application grid: simple at 2,500,000 to 4,000,000 FCFA (EUR 3,810 to 6,100) or medium at 4,000,000 to 7,000,000 FCFA (EUR 6,100 to 10,670), with maintenance from 35,000 to 120,000 FCFA a month. One requirement: data must stay hosted in the EU, under a contract that excludes training models on your documents.
Limits to understand before signing
An AI that writes a generic proposal loses points. Public buyers spot interchangeable text, and some tender rules now require bidders to disclose the use of generative tools. Human review is therefore not optional: every first draft is checked by the site manager who will run the job. The AI must never invent a reference, a headcount or a certification, so the system blocks any claim without an internal source.
Mini case study
Ciaran runs the bid office of a 140-employee structural contractor in Dublin. The team answers 120 tenders a year at 2.5 days per proposal, 300 working days. With the system, each proposal takes 1.25 days: 150 days saved per year.
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
At a fully loaded cost of EUR 450 per bid engineer day, the saving reaches 150 × 450 = EUR 67,500 a year. Year one costs EUR 20,000 of setup plus 12 × EUR 150 of API, EUR 21,800 in total. Payback comes in under 4 months. Ciaran chose to reinvest half the freed time into site-specific method statements, the section that actually wins points, and to bid on 20 more tenders a year.
FAQ
Can a public buyer penalise an AI-written proposal?
Buyers mostly penalise generic proposals. If the text cites your real projects, named team members and phasing adapted to the site, the score holds; some tender rules ask you to declare generative tool use, which takes one line.
How many past proposals do we need to start?
About 50 recent proposals are enough for a useful first result, but 200 to 600 documents covering 5 to 10 years give noticeably sharper drafts.
Are our documents exposed outside the company?
Not with a sound architecture: EU hosting, an API contract with no data reuse and role-based access. Dedicated European hosting stays under EUR 200 a month at this volume.
How long does setup take?
Plan 8 to 12 weeks, including 3 to 4 weeks to structure the archives. The first assisted proposals are produced from week six on a pilot scope.
Does it work for private tenders?
Yes, with adapted templates: private tenders weigh the proposal less and the price more, but the reference library also feeds commercial offers and prequalification files.
Let's scope your project. Send us your archive volume and the number of tenders you handle per year, and we will scope the library, the tool and an 8 to 12 week pilot. Detailed quote within 48 h. WhatsApp +221 77 596 93 33.
Mohamed Bah
Fondateur, Kolonell
Passionate about digital and entrepreneurship in Africa, Mohamed has been helping Sénégalese businesses with their digital transformation since 2020. Founder of Kolonell, he believes every SME deserves a professional and accessible online présence.
